ipv6: Handle route deletion notification

For the purpose of route offload, when a single route is deleted, it is
only of interest if it is the first route in the node or if it is
sibling to such a route.

In the first case, distinguish between several possibilities:

1. Route is the last route in the node. Emit a delete notification

2. Route is followed by a non-multipath route. Emit a replace
notification for the non-multipath route.

3. Route is followed by a multipath route. Emit a replace notification
for the multipath route.

In the second case, only emit a delete notification to ensure the route
is no longer used as a valid nexthop.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 43 insertions, 1 deletions

diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c
index 51cf848e38f0..67ddee539f77 100644
--- a/net/ipv6/ip6_fib.c
+++ b/net/ipv6/ip6_fib.c
@@ -415,6 +415,18 @@ int call_fib6_multipath_entry_notifiers(struct net *net,
 	return call_fib6_notifiers(net, event_type, &info.info);
 }
 
+int call_fib6_entry_notifiers_replace(struct net *net, struct fib6_info *rt)
+{
+	struct fib6_entry_notifier_info info = {
+		.rt = rt,
+		.nsiblings = rt->fib6_nsiblings,
+	};
+
+	rt->fib6_table->fib_seq++;
+	return call_fib6_notifiers(net, FIB_EVENT_ENTRY_REPLACE_TMP,
+				   &info.info);
+}
+
 struct fib6_dump_arg {
 	struct net *net;
 	struct notifier_block *nb;
@@ -1910,13 +1922,29 @@ static struct fib6_node *fib6_repair_tree(struct net *net,
 static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
 			   struct fib6_info __rcu **rtp, struct nl_info *info)
 {
+	struct fib6_info *leaf, *replace_rt = NULL;
 	struct fib6_walker *w;
 	struct fib6_info *rt = rcu_dereference_protected(*rtp,
 				    lockdep_is_held(&table->tb6_lock));
 	struct net *net = info->nl_net;
+	bool notify_del = false;
 
 	RT6_TRACE("fib6_del_route\n");
 
+	/* If the deleted route is the first in the node and it is not part of
+	 * a multipath route, then we need to replace it with the next route
+	 * in the node, if exists.
+	 */
+	leaf = rcu_dereference_protected(fn->leaf,
+					 lockdep_is_held(&table->tb6_lock));
+	if (leaf == rt && !rt->fib6_nsiblings) {
+		if (rcu_access_pointer(rt->fib6_next))
+			replace_rt = rcu_dereference_protected(rt->fib6_next,
+					    lockdep_is_held(&table->tb6_lock));
+		else
+			notify_del = true;
+	}
+
 	/* Unlink it */
 	*rtp = rt->fib6_next;
 	rt->fib6_node = NULL;
@@ -1934,6 +1962,14 @@ static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
 	if (rt->fib6_nsiblings) {
 		struct fib6_info *sibling, *next_sibling;
 
+		/* The route is deleted from a multipath route. If this
+		 * multipath route is the first route in the node, then we need
+		 * to emit a delete notification. Otherwise, we need to skip
+		 * the notification.
+		 */
+		if (rt->fib6_metric == leaf->fib6_metric &&
+		    rt6_qualify_for_ecmp(leaf))
+			notify_del = true;
 		list_for_each_entry_safe(sibling, next_sibling,
 					 &rt->fib6_siblings, fib6_siblings)
 			sibling->fib6_nsiblings--;
@@ -1969,8 +2005,14 @@ static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
 
 	fib6_purge_rt(rt, fn, net);
 
-	if (!info->skip_notify_kernel)
+	if (!info->skip_notify_kernel) {
+		if (notify_del)
+			call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL_TMP,
+						  rt, NULL);
+		else if (replace_rt)
+			call_fib6_entry_notifiers_replace(net, replace_rt);
 		call_fib6_entry_notifiers(net, FIB_EVENT_ENTRY_DEL, rt, NULL);
+	}
 	if (!info->skip_notify)
 		inet6_rt_notify(RTM_DELROUTE, rt, info, 0);